List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 2025 (Acc, ARI, ABO, AIRW)

80.6AccuracyStandard CoT (RL Final)
36.60848.02959.4570.871May 5, 2026
Updated 6mo ago

Evaluation results

MethodLinks
Standard CoT (RL Final)Backbone=Qwen3-30B-A3B, Training Stage=RL Final2026.0580.616,99516,70916,709
SxS Interleaved (RL Final)Backbone=Qwen3-4B, Training Stage=RL Final2026.058017,98117,8188,519
SxS Interleaved (RL Final)Backbone=Qwen3-30B-A3B, Training Stage=RL Final2026.0579.216,37516,32213,829
Standard CoT (SFT)Backbone=Qwen3-30B-A3B, Training Stage=SFT2026.0576.917,28716,95816,958
SxS Interleaved (RL Recovery)Backbone=Qwen3-30B-A3B, Training Stage=RL Recovery2026.057614,13214,0438,619
Base ModelBackbone=Qwen3-30B-A3B2026.057416,94216,94216,942
Standard CoT (RL Final)Backbone=Qwen3-4B, Training Stage=RL Final2026.0573.821,58021,31621,316
SxS Interleaved (RL Recovery)Backbone=Qwen3-4B, Training Stage=RL Recovery2026.0569.414,74414,9816,641
Base ModelBackbone=Qwen3-4B2026.0566.317,86217,33917,339
Standard CoT (SFT)Backbone=Qwen3-4B, Training Stage=SFT2026.0562.920,80720,54220,542
SxS Interleaved (SFT)Backbone=Qwen3-30B-A3B, Training Stage=SFT2026.0550.87,5408,8193,522
SxS Interleaved (SFT)Backbone=Qwen3-4B, Training Stage=SFT2026.0538.38,0029,6194,035